Message boards :
Number crunching :
seti, collatz and cuda
Message board moderation
Author | Message |
---|---|
dennis Send message Joined: 20 May 99 Posts: 2 Credit: 1,436,702 RAC: 3 |
I have been running seti cuda successfully for several months now. I upgraded to 6.10.43 about 3 weeks ago and all appeared to go well. A few days latter I attached to collatz and began cuda processing on that one. Ever since then seti cuda has not been working. I get seti work units, they say ready to process but never start. Collatz continues to crunch away finishing a cuda work unit every 3 hrs or so. I added debug statements to the xml file and am not getting any messages about too little gpu memory. Environment is vista 64b 6 gig, nvidia 9600m gt with 512 mem memory. Any ideas? Dennis |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
It could be because GPU tasks are done FIFO order, and you haven't got to the Seti Cuda tasks yet, or it could be because of Boinc 6.10.43, and the 197.xx drivers (which report less total RAM) Boinc 6.10.43 has a GPU work fetch bug, it'll report enough GPU RAM to get Cuda work, but not enough is available at run-time to actually run the application without crashing, Claggy |
Leopoldo Send message Joined: 4 Aug 99 Posts: 102 Credit: 3,051,091 RAC: 0 |
As addition to Claggy's message: Dennis, saw you scheduling values for this projects? (with project's "Properties" button in BOINC's "Advanced mode", search for "... scheduling priority" lines) Project with larger value should be crunched by BOINC sooner than its counterpart... (Except the case of projected crunching time is longer than task's deadline) ___________ WBW, Leonid |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Also, i notice up until 23 Apr you were running v6.08 (cuda) work, which only requires 254Mb GPU RAM, Did you upgrade your drivers as well at this point? after that you were getting v6.09 (cuda23) work which requires 384Mb GPU RAM, I noticed after i had upgraded to the 197.xx drivers, My 9800GTX+ 512Mb became 495Mb reported, and i could get v6.09 (cuda23) from Seti Beta, but couldn't start it because i only had 382Mb available when the tasks wanted to start, you could try Boinc 6.10.56 which is supposed to fix this, i think it'll just mean you'll only get v6.08 (cuda) work, and no more v6.09 (cuda23) work, or you could downgrade to 6.10.18, that version won't stop the v6.09 (cuda23) from starting, and will probably run fine, or you could try and free up a bit of GPU RAM by disabling some of Vista's Aero Effects, or you could get the optimised apps, which over-ride those memory restrictions. Claggy |
Gundolf Jahn Send message Joined: 19 Sep 00 Posts: 3184 Credit: 446,358 RAC: 0 |
I added debug statements to the xml file and am not getting any messages about too little gpu memory. Did you get any additional messages after setting the logging flags? Are you sure that the cc_config.xml is read correctly? Which flags did you set? Gruß, Gundolf |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
I have been running seti cuda successfully for several months now. I upgraded to 6.10.43 about 3 weeks ago and all appeared to go well. A few days latter I attached to collatz and began cuda processing on that one. Ever since then seti cuda has not been working. I get seti work units, they say ready to process but never start. Collatz continues to crunch away finishing a cuda work unit every 3 hrs or so. I added debug statements to the xml file and am not getting any messages about too little gpu memory. Environment is vista 64b 6 gig, nvidia 9600m gt with 512 mem memory. Any ideas? If you are going to use any version past 6.10.18 the only candidate that is without show-stopper issues is 6.10.56 at this time... even my old stand-by .45 had a GPU bug... it just took some time to surface ... So, either fall back to the current recommended, or leap forward and start from there ... |
dennis Send message Joined: 20 May 99 Posts: 2 Credit: 1,436,702 RAC: 3 |
<<As addition to Claggy's message: Dennis, saw you scheduling values for this projects? (with project's "Properties" button in BOINC's "Advanced mode", search for "... scheduling priority" lines) Project with larger value should be crunched by BOINC sooner than its counterpart... (Except the case of projected crunching time is longer than task's deadline)>> The collatz scheduling priority is roughly double that of seti, can that be changed? <<Also, i notice up until 23 Apr you were running v6.08 (cuda) work, which only requires 254Mb GPU RAM, Did you upgrade your drivers as well at this point? after that you were getting v6.09 (cuda23) work which requires 384Mb GPU RAM, I noticed after i had upgraded to the 197.xx drivers, My 9800GTX+ 512Mb became 495Mb reported, and i could get v6.09 (cuda23) from Seti Beta, but couldn't start it because i only had 382Mb available when the tasks wanted to start, you could try Boinc 6.10.56 which is supposed to fix this, i think it'll just mean you'll only get v6.08 (cuda) work, and no more v6.09 (cuda23) work, or you could downgrade to 6.10.18, that version won't stop the v6.09 (cuda23) from starting, and will probably run fine, or you could try and free up a bit of GPU RAM by disabling some of Vista's Aero Effects, or you could get the optimised apps, which over-ride those memory restrictions.>> I did upgrade my nvidia drivers to 197.16 after installing 6.10.43 and i dont used Aero graphics. The .56 version is not recommended yet, do you have any experience with it? <<Did you get any additional messages after setting the logging flags? Are you sure that the cc_config.xml is read correctly? Which flags did you set?>> Here is the contents of my config file and it was read successfully because I could see the messages and seti did not make any mention of gpu memory problems. <cc_config> <log_flags> <cpu_sched>1</cpu_sched> <cpu_sched_debug>1</cpu_sched_debug> </log_flags> </cc_config> Thanks for all your suggestions guys, I cant go back to 6.10.18 because Einstein wont run under that version. I guess my best bet is to try 6.10.53? |
Gundolf Jahn Send message Joined: 19 Sep 00 Posts: 3184 Credit: 446,358 RAC: 0 |
Here is the contents of my config file and it was read successfully because I could see the messages and seti did not make any mention of gpu memory problems. You could try to add <coproc_debug>1</coproc_debug> to see if that gives more useful info. Gruß, Gundolf |
Leopoldo Send message Joined: 4 Aug 99 Posts: 102 Credit: 3,051,091 RAC: 0 |
As addition to Claggy's message: Dennis, saw you scheduling values for this projects?The collatz scheduling priority is roughly double that of seti, can that be changed? Dennis, you can read article http://www.boinc-wiki.info/Work_Scheduler to better understanding BOINC principle of chosing the projects to crunch right now. * Work Scheduler does "round-robin" scheduling among Results, attempting to honor Resource Shares. * If I will wish to crunch 2 projects simultaneously, I will set equal Resource Share to both of them. Also, in your particular case of Collatz joining later than SETI, "Long Term Debt" for SETI is much lower than LTD for Collatz. Back to your question. To equalize BOINC expectations about projects with equal resource shares - there exist the way to change project debts, with "boinccmd" tool. But it does require advanced skills and caution. It's not recommended for majority of users - to use service tools for manual BOINC manipulating. I did it w/o problem. Backup BOINC data directory first (just for any case). At the BOINC directory exists service executable called "boinccmd.exe". To equalize debts for 2 projects, SETI and Collatz, you can run this executable with special parameters. Command line is: boinccmd --set_debts setiathome.berkeley.edu 0 0 boinc.thesonntags.com/collatz 0 0 After this you can check properties for these projects to see how numbers changed. Edit: but firstly look at coproc_debug output please. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
I cant go back to 6.10.18 because Einstein wont run under that version. I guess my best bet is to try 6.10.53? In a word ... no... If you are going to use a bleeding edge version past the latest "Recommended" the only choice now is 6.10.56 ... the only one in the whole 40-50s series that is almost as good is .45 and I no longer recommend it either ... So, to my mind you have two good choices at the moment... 6.10.18 or 6.10.56 ... as to experience, yes, 6.10.56 is what I am running on 5 systems ... and as of this morning there are no reported show stopper bugs. A couple of minor issues that I and others have reported have been fixed in the source and we should see in post .56 versions, but those are inconsequential and do not affect actual operation. |
Tony Li Send message Joined: 21 May 01 Posts: 6 Credit: 1,337,747 RAC: 0 |
Are other folks currently getting cuda units? I've had cuda enabled for awhile now, and get cuda work from Collatz, but not from Seti. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Tony, Seti doesn't have a CUDA application for the Mac. Only for Windows. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.